A CHARACTERIZATION OF SOME RANK 2 INCIDENCE GEOMETRIES BY THEIR AUTOMORPHISM GROUP

Transcription

1 A CHARACTERIZATION OF SOME RANK 2 INCIDENCE GEOMETRIES BY THEIR AUTOMORPHISM GROUP F. Buekenhout Université Libre de Bruxelles Campus Plaine 216, B 1050 Bruxelles fbueken@ulb.ac.be H. Van Maldeghem Universiteit Gent Seminarie voor Meetkunde en Kombinatoriek Krijgslaan 281, B 9000 Gent hvm@cage.rug.ac.be December 14, 2010 Abstract Using the classification of the finite simple groups, we classify all finite pointline geometries with a diameter exceeding the gonality by at most 1 and having an automorphism group acting transitively on the set of maximal geodesics of each given type. Research Associate of the National Fund for Scientific Research (Belgium) 1

2 1 INTRODUCTION. 1.1 History and Motivation. Efforts in order to classify rank 2 geometries (bipartite graphs) with large automorphism groups have been made in various contents such as the Moufang polygons (see Tits [85, 86, 87, 88, 89], Weiss [94], Thas, Payne & Van Maldeghem [82]), distance transitive graphs (Brouwer, Cohen & Neumaier [10]), distance transitive generalized polygons (Buekenhout & Van Maldeghem [15, 16]) flag-transitive designs (Kantor [58], Buekenhout, De Landtsheer, Doyen, Kleidman, Liebeck & Saxl [14]). A synthesis and further deepening of these efforts seems suitable and possible. The present paper takes this direction. In particular, we want to generalized the results of [15, 16] on generalized polygons to a larger class of geometries. We consider a rank 2 geometry Γ which is a (g, d p, d l )-gon with 2 g d p d l g + 1 (see 1.2.1). That situation includes the generalized polygons, the linear spaces, the partial geometries, the Moore geometries and the symmetric 2-designs. Some very interesting geometries escape to it, such as partial geometries. We assume that Γ is finite and that it is equipped with an automorphism group acting transitively on the ordered maximal geodesics of each possible type. This is a weakening of the Moufang condition and so we get somewhat better results in that direction (our proof however uses the classification of the finite simple groups). On the other hand, our condition is much stronger than the flag-transitivity condition used successfully in some of the earlier work. We get a complete classification. This provides an objective basis for classes of geometries enlarging the class of classical generalized polygons. It may be useful for extensions to geometries of rank greater than or equal to three. 1.2 Definitions and Notation Regular (g, d p, d l )-gons. A rank 2 point-line incidence geometry Γ consists of a triple (P, L, I), where P is the set of points, L is the set of lines and I (P L) (L P) is a symmetric incidence relation. The elements of P and L are also called varieties and the type typ(x) of a variety is its name (the appropriate point or line ). The elements of I are usually called flags. Two points incident with some line are called collinear and two lines incident with some point are concurrent. A path γ of length n based at a variety x is a sequence (x = x 0, x 1, x 2,..., x n ) of n + 1 varieties with x i 1 Ix i for 1 i n. If x 0 = x n and if x i x i+2 (for all i to be taken modulo n), then n is even and γ is called a circuit of diameter n/2. We call Γ connected if every two varieties can be joined by a path. The distance d(x, y) between two varieties x and y is the length of the shortest path joining x to y (well defined by connectedness). A geodesic (based at) x is a path γ based at x such that the length of γ is equal to the distance between the extremeties of γ. A maximal geodesic is a geodesic that is not properly contained in another one. The gonality g of Γ is the diameter of the smallest circuit (i.e. a circuit of minimal diameter) in Γ. The local diameter d(x) of some variety x is the length of the longest geodesic based at x. The 2

3 maximal value of d(x) for x a point (resp. a line) is the point- (resp. line-) diameter and it is denoted by d p (resp. d l ). With this notation, Γ is called a (g, d p, d l )-gon. The dual Γ D = (L, P, I) of Γ is obviously a (g, d l, d p )-gon (see Buekenhout [12], where this notion is introduced). Let x be any variety of Γ. Then we denote by Γ i (x) the set of varieties at distance i from x. We call the geometry Γ regular if Γ i (x) Γ j (y) = Γ i (z) Γ j (u) for all positive integers i, j and all varieties x, y, z, u whenever d(x, y) = d(z, u) and typ(x) =typ(z) (and hence typ(y) =typ(u)). In particular putting i = 1, j = 2 and d(x, y) = 1 (i.e. x and y are incident), we see that in a regular geometry Γ the number of points (resp. lines) incident with a given line (resp. point) is a constant, say s+1 (resp. t+1). In that case, we call (s, t) the order of Γ. If s > 1 and t > 1, then we say that Γ is thick (terminology of buildings, see e.g. Tits [84]). Also, it is straightforward to see that in a regular point-line geometry Γ the length of a maximal geodesic only depends on the types of its extremeties, in other words, the local diameter in every point (resp. line) x is equal to the point-diameter (resp. line-diameter). Note that if d p = d l and this is even, then there are two types of maximal geodesics: one kind has points as extremeties and the other kind has lines. In any case we trivially have the inequality d p d l 1 and by point-line duality, we may assume g d p d l, in other words we assume that the diameter (which is in general the larger value among d p, d l ) is equal to d l. Note that if d l is odd, then d p = d l (this is obvious, see also Buekenhout [12]). Finally, a graph is a geometry in which every line is incident with exactly 2 points. In this case, the points are called vertices (adjacent if they are collinear) and the lines edges. The incidence graph of an arbitrary geometry Γ = (P, L, I) is the graph whose set of vertices is P L and in which two vertices form an edge if they form a flag in Γ. We denote the incidence graph of Γ by Γ I. In this paper, we will always assume that Γ is finite, connected and regular Some classical examples. The following particular cases provide the main motivation for (g, d p, d l )-gons. In most cases that we mention g d p d l g + 1. A generalized n-gon is a regular (n, n, n)-gon and conversely. These were introduced by Tits [83]. If n = 3, they are projective planes. For n = 4, 6, 8, they are called generalized quadrangles, respectively generalized hexagons,generalized octagons. A generalized 2-gon (or generalized digon) is a trivial incidence geometry (every point is incident with every line). By a theorem of Feit & Higman [33], a generalized n-gon of order (s, t) with s, t 2 can only exist if n {2, 3, 4, 6, 8}. A linear space is any geometry with gonality 3 and point-diameter 3, hence in a linear space, two points determine uniquely a line. So a linear space is either a generalized projective plane (i.e. a (3, 3, 3)-gon) or a (3, 3, 4)-gon. A partial geometry with parameters (s, t, a), as introduced by Bose [6], is a pointline geometry Γ = (P, L, I) of order (s, t) (defined as above) having the properties that (1) 3

4 every point x is collinear to exactly a + 1 points incident with any given line not incident with x and (2) two points determine at most one line. Hence the diameter is at most 4 and we have the following possibilities: 1. Γ is a regular (3, 3, 3)-gon, i.e. a projective plane, so Γ has parameters (s, s, s) for some positive integer s. 2. Γ is a regular (3, 3, 4)-gon or a regular (3, 4, 3)-gon, i.e. Γ is a regular proper linear space or a regular proper dual linear space. 3. Γ is a regular (3, 4, 4)-gon. Among these, we have the nets and the dual nets (see later). The other members in this class are called the proper partial geometries. 4. Γ is a generalized quadrangle and has parameters (s, t, 0) for some positive integers s, t. A net of order a and degree b is a partial geometry with parameters (a 1, b 1, b 2). If a = b, then it has been called a helicopter plane in Van Maldeghem [93]; it is an affine plane with one parallel class of lines removed. A partial quadrangle, as introduced by Cameron [17], with parameters (s, t, a) is a point-line geometry Γ = (P, L, I) of gonality g 4 such that every two non-collinear points are collinear with exactly a + 1 points. In general, this is a (4, 5, 6)-gon, but if a = t, then we have a generalized quadrangle; if a = 0, then we have a (5, 5, 6) or (5, 5, 5)-gon; if Γ is also a dual partial quadrangle, then it is a regular (4, 5, 5)-gon. Partial quadrangles which are not generalized quadrangles are also known as near pentagons, see e.g. Brouwer, Cohen & Neumaier [10]. A Moore geometry is a (g, g, g +1)-gon for g 3 and g odd (see Buekenhout [12]). As for generalized polygons, there is here a restriction on g, see subsection A Moore geometry was originally defined as a point-line geometry of order (s, t) such that every two points are joined by a unique geodesic, see Bose & Dowling [7]. A symmetric design or square design with parameters (v, k, λ) is an incidence structure Γ = (P, L, I) of order (k 1, k 1) such that 2 blocks intersect in λ points and 2 distinct points determine exactly λ blocks. The positive integer v is the total number of points. Γ is also called a 2 (v, k, λ)-design. If 1 < λ < k, then Γ is a regular (2, 3, 3)-gon. If λ = k, it is a generalized digon and if λ = 1, then Γ is a projective plane, hence a regular (3, 3, 3)-gon. The complementary design of a symmetric design Γ is the symmetric design Γ C obtained from Γ by replacing each block by its complement Automorphisms, collineations and correlations. A collineation of the geometry Γ = (P, L, I) is a permutation on P L preserving P, L and I. A correlation is a permutation on P L preserving I and interchanging P and L. An automorphism is either a collineation or a correlation. 4

5 In the same way one defines isomorphisms and anti-isomorphisms between geometries and of course we are only interested in the isomorphism class of a geometry. If Γ is anti-isomorphic to itself, i.e. there is a correlation of Γ, then we call Γ self-dual. The group of all collineations, resp. automorphisms of Γ will be denoted by Col(Γ), resp. Aut(Γ). Now let G be an automorphism group of the geometry Γ and suppose Γ is a (g, d p, d l )- gon. We shall use the following terminology. 1. Suppose G acts transitively on the set of pairs (x, y) of points at distance i from each other, for all even positive integers i, then we call (Γ, G) a point distance transitive (g, d p, d l )-pair, dually a line distance transitive (g, d p, d l )-pair. If (Γ, G) is both point distance transitive and line distance transitive, then we call (Γ, G) a weakly distance transitive (g, d p, d l )-pair. If G acts transitively on each set of pairs of varieties at distance j from each other and having fixed type, for all positive integers j, then (Γ, G) is called a distance transitive (g, d p, d l )-pair. Moreover if G contains moreover a correlation, then (Γ, G) is a full distance transitive (g, d p, d l )-pair. 2. Suppose G acts transitively on each set of geodesics based at some point x of Γ and ending in a point y at maximal distance, for all points x P, then we call (Γ, G) a point geodesic transitive (g, d p, d l )-pair. Similarly as for distance transitive (g, d p, d l )-pairs, we can define line geodesic transitive (g, d p, d l )-pairs, respectively weakly geodesic transitive, geodesic transitive and full geodesic transitive (g, d p, d l )-pairs. 3. If G acts transitively on each set of geodesics of length i based at some fixed variety x, for all varieties x, then (Γ, G) is called a locally i-arc transitive (g, d p, d l )-pair. If G acts transitively on the full set of geodesics of length i, then (Γ, G) is called i-arc transitive. This generalizes a similar notion for graphs, see e.g. Weiss [95], as it was first introduced by Tutte [91]. It is easy to see that, if Γ is a (g, d p, d l )-gon and if 2 g d p d l g + 1, then each of the above assumptions on G implies that Γ is regular. Hence from now on we assume that all geometries are finite and regular unless the contrary is explicitly mentioned. 1.3 Main Results. Now we are ready to formulate our main results and some immediate corollaries. For a description of the geometries and groups under consideration, we refer to section 2, in particular, we write appropriate automorphism group in order not to overload the formulation, but all these groups are described in section 2. The proof of theorem 1 follows from propositions 1 up to 8 of section 4. The symbol q will always denote a prime power. For the notation of groups, we follow the Atlas [22]. THEOREM 1. Let (Γ, G) be a finite geodesic transitive (g, d p, d l )-pair, 2 g d p d l g + 1, then one of the following holds: 5

6 1. Γ is a thick generalized polygon related to an irreducible finite adjoint or twisted adjoint Chevalley group X n (q) of relative rank 2 and X n (q) G Aut(X n (q)), or Γ is the flag complex of the self-dual thick generalized polygon related to L 3 (q), S 4 (2 e ) or G 2 (3 e ) and G is as above with the additional condition that it contains a graph automorphism, or G = A 6 and Γ is the unique generalized quadrangle of order (2, 2), or Γ is an ordinary polygon and G is the corresponding dihedral group; 2. Γ can be identified with the Petersen graph on 10 points, resp. Hoffman-Singleton graph on 50 points; the lines are the edges of the graph and G = S 5, resp. U 3 (5) G U 3 (5) : 2. Here, Γ can be considered as a Moore geometry, in particular as a (5, 5, 6)- gon of order (1, 2), resp. (1, 6); 3. Γ is a (3, 4, 4)-gon. The following cases occur: 3.1. Γ is the helicopter plane HG(2, q) obtained from the Desarguesian projective plane P G(2, q) by deleting a flag (x, l) and all varieties incident with one of x, l and G contains the stabilizer in P GL 3 (q) of the flag (x, l) in P G(2, q); 3.2. Γ is the net (H n+1 q ) D of order q n and degree q + 1 and G contains a group isomorphic to the semi-direct product of an elementary abelian group q 2n with a group isomorphic to (a) (SL 2 (q) SL n (q))/z(sl 2 (q) SL n (q)) if n > 2, or (b) (SL 2 (q) GL 2 (q))/z(sl 2 (q) GL 2 (q)) if n = 2, or (c) SL 2 (2) A 7 if (n, q) = (4, 2) Γ is the dual of 3.2.; 3.4. Γ is the net Ne(2 8 ): its points can be identified with the points of an affine space AG(8, 2) and its lines are the affine 4-subspaces whose 3-spaces at infinity constitute a 2-transitive spread of a hyperbolic quadric in P G(7, 2); G contains the full translation group of AG(8, 2) and its kernel at infinity is A 9 ; 3.5. Γ is the dual of Γ is a (3, 3, 4)-gon. Three cases occur: 4.1. Γ is the linear space consisting of the points and lines of P G(d, q), q 3 and L d+1 (q) G P ΓL d+1 (q); 4.2. Γ is the affine Desarguesian plane AG(2, q), G contains all translations and induces at the line at infinity a group containing L 2 (q) G is a group acting 4-transitively on the set of points of Γ and the lines of Γ can be identified with the pairs of points; 5. Γ is a (2, 3, 3)-gon. Here, Γ is a symmetric 2-design with λ > 1 and four cases occur: 6

7 5.1. Γ can be identified with P G(d, q), d 3, the blocks are either the hyperplanes or their complements and L d+1 (q) G P ΓL d+1 (q) : 2 or G = A 7, S 7 (if (d, q) = (3, 2) and blocks are the hyperplanes); 5.2. Γ is the Paley (or Hadamard) design on 11 points and L 2 (11) G L 2 (11) : 2; 5.3. Γ is isomorphic to one of Kantor s designs S ± (n) and G = 2 2n : Sp 2n (2); 5.4. G acts 3-transitively on the set of points of Γ and the blocks are the complements of the points; 6. Γ is a generalized quadrangle of order (1, s) or (s, 1) and G is appropriate; 7. Γ is a generalized digon. An immediate corollary is the following: COROLLARY 1. Let (Γ, G) be a full geodesic transitive (g, d p, d l )-pair, 2 g d p d l g + 1, then one of the following holds: 1.1. Γ is the generalized quadrangle of order (q, q) appearing in conclusion 1 of theorem 1 for every even q and G is the appropriate group containing a correlation; 1.2. Γ is the generalized hexagon of order (q, q) appearing in conclusion 1 of theorem 1 for every q divisible by 3 and G is the appropriate group containing a correlation; 2 Γ is the helicopter plane HG(2, q) as in conclusion 3 of theorem 1 and G is appropriate but containing a correlation; 3 (Γ, G) is as in conclusion 5 of theorem 1 (case of symmetric 2-designs) with the only restriction that G contains a correlation; 4 Γ is generalized digon. For various sub-classes of geometries, we obtain more general results by weakening the hypothesis on G. We refer to section 4 for the precise statements. As a byproduct of our proof, we obtain a result on partial quadrangles (and they do not necessarily satisfy d l g + 1, so they are not covered by theorem 1), see also section 4, proposition 4 for a more detailed statement and the proof. THEOREM 2. Let (Γ, G) be a point geodesic transitive (g, d p, d l )-pair with Γ a partial quadrangle of order (s, t). Then one of the following possibilities occur: 1. Γ is a generalized quadrangle (and (Γ, G) is one of the examples in the conclusion of theorem 2); 7

8 2. s = 1 and Γ is one of the following graphs: the pentagon, Petersen, Clebsch, Hoffman- Singleton, Higman-Sims on 100 (resp. 77) vertices, Gewirtz. The group G is an appropriate automorphism group containing respectively D 10, A 5, 2 4 : (5 : 4), U 3 (5), HS, M 22, L 3 (4); 3. Γ is a partial quadrangle with parameters (2, 10, 1) constructed in AG(5, 3) and G is appropriate; 4. Γ is isomorphic to the partial quadrangle T 3 (O) with O an elliptic quadric or the Suzuki-Tits ovoid in P G(3, q) and G is appropriate. In cases 3 and 4, G acts on an affine space, contains all translations and the stabilizer of a point contains a normal subgroup isomorphic to one of M 11, L 2 (q 2 ), Sz(q). 8

9 2 THE EXAMPLES. In this section, we give all the examples mentioned in theorems 1 and 2 and proposition 1 below and we briefly comment the properties of their automorphism group. 2.1 Generalized Polygons The Moufang generalized polygons. The classical examples (i.e. those which have the Moufang property, see above and section 5) of generalized polygons, due to Tits [83], arise from Chevalley groups. We give a brief description. Let G = X n (q) be a Chevalley group, q = p h where p is a prime and let B be the normalizer of a Sylow p-group in G (B is called a Borel-subgroup). If G is one of the groups of table 1, there are exactly two maximal subgroups containing B, denote them by P 1 and P 2 ; they are called the maximal parabolic subgroups of the pair (G, B). We define a geometry Γ = (P, B, I) as follows: the points are the left cosets of P 1 in G and the lines are the left cosets of P 2 in G, a point and a line being incident exactly when the corresponding cosets are not disjoint. The geometry Γ thus obtained is a thick generalized n-gon of order (s, t) as listed in table 1 (where the set of points is chosen in the usual classical way). In every case the pair (Γ, G) has both the Moufang and the Tits property and every group G Aut(Γ) acting point distance transitively on Γ contains G (for the right choice of the points in the above construction), except if G is not simple, i.e. if (1) G = S 4 (2) = O 5 (2), (2) G = G 2 (2) and (3) G = 2 F 4 (2) = R(2); in these cases also the derived group G acts point distance transitively on Γ. In case (1), G acts distance transitively on Γ, but (Γ, G ) does not have the Tits property nor the Moufang property, it is not even half Moufang. In cases (2) and (3), G does not act line distance transitively and consequently it does not induce the Tits nor the Moufang property; but it does induce the half Moufang property. Note that for a generalized n-gon Γ, n even, the pair (Γ, G) is distance transitive if and only if it is special distance transitive. This is an immediate consequence of the definition. G n (s, t) Remarks (GP1) L 3 (q) 3 (q, q) Self-dual (GP2) S 4 (q) 4 (q, q) Self-dual iff q is even (GP3) O 5 (q) 4 (q, q) Dual to (GP2) (GP4) O6 (q) 4 (q, q 2 ) (GP5) U 4 (q) 4 (q 2, q) Dual to (GP4) (GP6) U 5 (q) 4 (q 2, q 3 ) (GP7) G 2 (q) 6 (q, q) Self-dual iff q is a power of 3 (GP8) 3 D 4 (q) 6 (q, q 3 ) (GP9) 2 F 4 (q) 8 (q, q 2 ) q is an odd power of 2 Table 1: Finite Thick Moufang Generalized n-gons. 9

10 The incidence graph Γ I of a generalized n-gon Γ of order (q, q) is a generalized 2n-gon of order (1, q). If Γ is Moufang, then also Γ I is Moufang, but Γ I has the Tits property, respectively is distance transitive, geodesic transitive only if Γ has the Tits property and is self-dual, respectively is full distance transitive, full geodesic transitive The unique generalized quadrangle of order (3, 5). This example is one out of a class of generalized quadrangles of order (s, s + 2) due to Ahrens & Szekeres [1]. Consider the projective plane P G(2, 4) and a complete oval O in it, i.e. a conic together with its kernel. Embed P G(2, 4) as a hyperplane in P G(3, 4) and define the following geometry Γ = (P, L, I): the elements of P are the points of P G(3, 4) not in P G(2, 4); the elements of L are the lines in P G(3, 4) meeting O in exactly 1 point; incidence is the natural one. Then Γ is a generalized quadrangle of order (3, 5) and it is usually denoted by T2 (O). For more information on this interesting quadrangle we refer to a recent paper of Payne [73]. We just mention that the full collineation group of T2 (O) contains all translations and homologies of P G(3, 4) leaving P G(2, 4) pointwise invariant and its kernel on P G(2, 4) is the full automorphism group of O which is S 6, the full symmetric group acting in its standard representation on the six points of O. So Col(T2 (O)) acts point distance transitively on T2 (O), but obviously not line distance transitively Non-Thick Generalized Quadrangles. Consider a generalized quadrangle Γ = (P, L, I) of order (s, 1). This is actually just an (s + 1) (s + 1)-grid. Consider a group G acting transitively on that grid and such that the stabilizer G x of any point x acts transitively on both Γ 2 (x) and Γ 4 (x). Then G acts geodesic transitively on Γ. So point distance transitivity implies geodesic transitivity. This case, and the dual one, corresponds to case 6 of theorem 1. The corresponding groups are described in subsection 3.2, class III of the permutation rank 3 groups. 2.2 Partial Quadrangles Thick partial quadrangles. Consider the projective 3-space P G(3, q) and an ovoid O in it (an ovoid in P G(3, q) is a set of q points no three of which are collinear). Embed P G(3, q) as a hyperplane in P G(4, q) and define the following geometry T 3 (O): the points are the points of P G(4, q) not in P G(3, q) and the lines are the lines of P G(4, q) meeting P G(3, q) in a single point x of O. Then T 3 (O) is a partial quadrangle with parameters (q 1, q 2, q 2 q 1). There are two cases for which the collineation group of T 3 (O) acts point distance transitively on T 3 (O): 1. The first case is when O is a non-ruled quadric (an elliptic quadric), i.e. a set of points satisfying the equation X 0 X 1 = f(x 2, X 3 ), where f(x, y) is an irreducible quadratic form over the field GF (q). 10

11 2. The second case is when O is the Suzuki-Tits ovoid for spaces over GF (q) with q = 2 2e+1. The points of O can be described in coordinates as {(x, y, xy + x s+2 + y s, 1)} {(0, 0, 1, 0)}, s = 2 e+1. The case where O is a quadric can also be constructed from the generalized quadrangle (GP4) by deleting all points collinear with one fixed point and deleting all lines through that point (see Cameron [17]). In fact, every partial quadrangle with parameters (q 1, q 2, q 2 q 1) can be constructed that way as proved by Ivanov & Shpectorov [53]. In subsection 4.2, we show that the two examples above are in fact point geodesic transitive. Consider now P G(4, 3), there is a cap C in P G(4, 3) consisting of 11 points and the Mathieu group M 11 acts on C in its standard 4-transitive action (a cap is a set of points no three of which are collinear). Repeating the construction from the previous paragraph (substituting C for O, P G(4, 3) for P G(3, q), etc... ), one obtains a partial quadrangle with parameters (2, 10, 1) which we shall denote by Hi(243), see Hill [47] and Berlekamp, van Lint & Seidel [4]. The above examples are constructed as linear representations, i.e. the points of the geometry Γ are all points of a certain affine space A and the set of lines of Γ is a union of parallel classes of lines of A Strongly regular graphs with λ = 0. A graph is called regular if every vertex has a constant number k of adjacent vertices (adjacent vertices are vertices on one edge). A regular graph is called strongly regular if every two adjacent vertices are both adjacent to a constant number λ of vertices and if every two non-adjacent vertices are both adjacent to a constant number µ of vertices. In this case the strongly regular graph is said to have parameters (v, k, λ, µ), where v is the total number of vertices. Note that, if Γ is a strongly regular graph with parameters (v, k, λ, µ), then the graph obtained by interchanging edges with non-edges is again a strongly regular graph, called the complementary strongly regular graph. If we denote its parameters by (v, l, λ, µ), then we have the relations: k + l = v; λ = l k + µ 1; µ = l k + λ + 1; and µl = k(k λ 1). For any strongly regular graph, we will use this standard notation without further comments (see Hubaut [48]). If Γ is a strongly regular graph with λ = 0, i.e. Γ does not contain triangles, then it is a partial quadrangle with parameters (1, k 1, µ 1). Now suppose a group G has a rank 3 permutation representation on a set Γ (for the definition see section 3 below) and let G x be the stabilizer of an element x of Γ. Let x i, 11

12 i = 1, 2, be an element in the orbit i of G x (where we assume that the three orbits are numbered 0,1,2 and orbit 0 is the trivial one). Then we define edges in Γ by letting G act on {x, x i }. If G is even, then this defines two mutually complementary strongly regular graphs (see Higman [45]). Table 2 contains a list, taken from Hubaut [48], of some strongly regular graphs without triangles constructed in this way, where G is a simple group. The table contains the label for future reference, the group G, the subgroup G x, the notation for γ (and we make the convention that we choose x i in the smallest suborbit, with the above notation), the parameters of Γ as a strongly regular graph and the name in the literature. G G x Γ (v, k, λ, µ) Name (PQ0) D 10 2 P n(5) (5,2,0,1) Pentagon (PQ1) A 5 S 3 P e(10) (10,3,0,1) Petersen (PQ2) U 3 (5) A 7 HS(50) (50,7,0,1) Hoffman-Singleton (PQ3) L 3 (4) A 6 Ge(56) (56,10,0,2) Gewirtz (PQ4) M : A 6 HS(77) (77,16,0,4) Higman-Sims (PQ5) HS M 22 HS(100) (100,22,0,6) Higman-Sims Table 2: Rank 3 Graphs with λ = 0 related to Simple Groups. We need one further non-trivial strongly regular graph without triangles underlying a rank 3 group, namely the Clebsch graph Cl(16). There are several descriptions of it and here is a less usual one: the vertices of Cl(16) are the elements of the field GF (16) of 16 elements and two vertices form an edge if their difference in GF (16) is a third power. The collineation group of Cl(16) is isomorphic to 2 4 : S 5, but we already have a rank 3 representation if we take the subgroup 2 4 : D 10, where G x = D10 is generated by the multiplication with third powers in GF (16) and the involutory automorphism of GF (16), taking x = 0. The parameters of Cl(16) are (16, 5, 0, 2) and we label this example (PQ6). In subsection 4.2, we determine the transitivity properties of the above mentioned strongly regular graphs (see table 20). 2.3 Some Nets Helicopter planes. Consider an affine plane of order q (i.e. there are q points on each line; with our convention, it has order (q 1, q)) and delete one entire class of parallel lines. The incidence geometry Γ thus obtained is a net of order q and degree q. Suppose the original affine plane was the Desarguesian plane AG(2, q) and consider the subgroup G of P GL(2, q) stabilizing Γ. Then (Γ, G) is a geodesic transitive (3, 4, 4)-pair (for q 3). 12

13 2.3.2 The net (H n q ) D. Consider the following geometry Hq n : the points are the points of the projective space P G(n, q) which are not contained in a fixed subspace P G(n 2, q) of P G(n, q); the lines are the lines of P G(n, q) disjoint from P G(n 2, q); incidence is the natural one. This yields a dual net with parameters (q, q n 1 1, q 1). The corresponding net can be constructed in another way as follows: Consider a vector space V 2 resp. W n 1 of dimension 2 resp. n 1 over GF (q). The points of (Hq n ) D are the elements of the tensor product vector space V 2 Wn 1 and the lines are the sets of the form v W n 1 and its translates, where v is an arbitrary vector in V 2. By De Clerck & Johnson [27], theorem 4, this constitutes indeed the dual of Hq n. If a group G acts geodesic transitive on (Hq n ) D, then it is clear that the group induced on W n 1 by the stabilizer G o of the zero vector in V 2 Wn 1 acts 2-transitively on the vector lines of W n 1 (for the definition of 2-transitive group see subsection 3.1) and hence is known (see again subsection 3.1, in particular table 4) The net Ne(2 8 ). Consider the hyperbolic quadric in P G(7, 2) and a 2-transitive ovoid (see e.g. Kleidman [60]) in it. Apply triality to obtain a 2-transitive spread S. Embed P G(7, 2) as a hyperplane in P G(8, 2). Define as the point set of Ne(2 8 ) the set of points in P G(8, 2) \ P G(7, 2). A line is a 4-subspace of P G(8, 2) meeting P G(7, 2) in a member of S. This constitutes a net Ne(2 8 ) with parameters (15, 8, 7). The automorphism group of Ne(2 8 ) is isomorphic to 2 8 : A 9, where A 9 is the group acting 2-transitively on the elements of the spread. 2.4 Linear Spaces. Here, we simply list in table 3 all linear spaces which admit a flag-transitive collineation group (and have s, t 1). This result is due to Buekenhout, De Landtsheer, Doyen, Kleidman, Liebeck & Saxl [14]. We will give more information in the proof (paragraph 4.3.2). In the table, a c-geometry is a geometry in which the lines are all pairs of points. Linear space s t (q n q) (q 1) (LS1) P G(n, q), n 2 q (LS2) Hermitian unital in P G(2, q 2 ) q q 2 1 (LS3) Ree unital arising from R(q), q = 3 h, h odd q q 2 1 (q 2) (LS4) Witt-Bose-Shrikhande space (defined for q even) q 2 (q (LS5) AG(n, q), n 2 q 1 q) (q 1) (LS6) Some non-desarguesian translation affine planes q 1 q (LS7) Hering spaces 8 90 (LS8) Affine line spaces (LS9) c-geometry on v points 1 v 2 Table 3: Linear Spaces Admitting a Flag-transitive Group. 13

14 Note that any set S defines a unique c-geometry in the obvious way. We denote that linear space by Γ(S). 2.5 Symmetric 2-designs. We mention some symmetric 2-designs for which the collineation group acts 2-transitively on the points; the classification of all such designs is due to Kantor [58]. (SD1) Desarguesian projective spaces. The blocks are the hyperplanes or the complement of the hyperplanes. The collineation groups are the linear or semi-linear groups or A 7 for P G(3, 2); (SD2) The Paley (or Hadamard) design on 11 points, denoted by Ha(11), is a 2-(11,6,2) design. The points are the elements of Z (mod 11) and the lines are the translates of {1, 3, 4, 5, 9}. Its collineation group is L 2 (11). The complementary design is a 2-(11,6,3) design; (SD3) The geometry of Higman on 176 points. This is a 2-(176,50,14) design with collineation group HS, the sporadic Higman-Sims group. We denote this design by Hi(176). The complementary design is a 2-(176,126,90) design; (SD4) A 2-(2 2n, 2 n 1 (2 n 1), 2 n 1 (2 n 1 1))-design S + (n) of which there is exactly one for each n 2, see Kantor [55]. The complementary design is denoted by S (n). The collineation group of both these designs is a group isomorphic to 2 2n : S 2n (2); (SD5) Any set Ω with a 2-transitive group acting on it can be turned into a 2-transitive symmetric 2-design by declaring the blocks to be the complements of the points. This can be defined for any set Ω without a 2-transitive group and we denote the corresponding design by Γ(.Ω). 14

15 3 PRELIMINARIES. The proofs of our main results are basically geometric in nature. But we make use of some major group-theoretical results such as the classification of all doubly transitive finite groups (Cameron [18] and Hering [44], the determination of all primitive rank 3 representations of finite groups (Kantor & Liebler [59], Bannai [3], Liebeck & Saxl [67] and Liebeck [65]), the classification of various classes of distance transitive graphs (Ivanov [52], Liebeck, Praeger & Saxl [66], Praeger, Saxl & van Bon [75] and Praeger & Soicher [76]), the determination of all large maximal subgroups of the exceptional groups (Liebeck [68]) and the enumeration of all primitive rank 4 and 5 representations of the Chevalley groups (Cuypers [23]). We now list these results (and some other) for future reference. We use the classification of the finite simple groups. They fall into five distinct (though non-disjoint) classes: 1. The cyclic groups of prime order; 2. The alternating groups A n for n 5; 3. The classical Chevalley groups, i.e. the linear, symplectic, orthogonal and unitary (simple) groups; 4. The exceptional Chevalley groups (including the Tits group here); 5. The 26 sporadic groups. The first chapters of the Atlas [22] contain an introduction to this subject. 3.1 Permutation Groups. Let Ω be a set and G a group acting faithfully on Ω. Then the pair (Ω, G) is said to have permutation rank n, n > 1, if G is transitive on Ω and if the stabilizer G x of some element x of Ω has exactly n orbits in Ω. A rank 2 group is also called a 2-transitive group. The group G acts n-transitively, n > 2 on Ω if G acts transitively on Ω and G x acts (n 1)-transitively on Ω \ {x}, for some x Ω. If Ω is an affine space and G contains the full translation group of Ω and is itself contained in the full automorphism group of Ω as an affine space, then we say that (Ω, G) is of affine type. If there is a non-abelian simple group S such that S G Aut(G), then we say that G is almost simple and (Ω, G) is of almost simple type. In this case the group S is the socle of G, denoted by Soc(G) (special case of a more general definition: Soc(G) is the subgroup of G generated by all minimal normal subgroups of G). A subset A Ω is a set of imprimitivity for (Ω, G) if A θ A is either empty or A itself, for all θ G. Note that we use exponential notation for the action of G on Ω. The action of G on Ω is called primitive if the only sets of imprimitivity are the trivial ones, i.e. the singletons and Ω itself. If (Ω, G) is 2-transitive, then it is automatically primitive 15

16 (this is easy to see). As a result of the classification of the finite simple groups, all finite 2-transitive groups are known. They are divided into two classes: the almost simple and the affine type. In table 4, we list all 2-transitive groups of almost simple type acting on a set Ω (see e.g. Cameron [18]) and label them for future reference. We list the minimal group; all other groups are obtained by adjoining group automorphisms. G Ω Restrictions and Remarks (TS1) A n n symbols n 5 (q (TS2) L n (q) 1) points of P G(n 1, q) (q 1) (d, q) (2, 2), (2, 3), 3-transitive if P GL 2 (q) G (TS3) U 3 (q) q points of a Hermitian unital q 3 (TS4) R(q) q points of the Ree unital q = 3 h, h 1 odd (TS5) Sz(q) q points of The Suzuki Ovoid q = 2 h, h 3 odd (TS6) S 2d (2) 2 2d 1 ± 2 d 1 non-degenerate quadrics d 3 (TS7) L 2 (11) 11 points of Ha(11) (TS8) A 7 15 points of P G(3, 2) (TS9) M points of a Steiner system 4-transitive (TS10) M points of a 3-design (TS11) M points of a Steiner system 5-transitive (TS12) M points of a Steiner system 3-transitive (TS13) M points of a Steiner system 4-transitive (TS14) M points of a Steiner system 5-transitive (TS15) HS 176 points of Hi(176) (TS16) Co points in the Leech lattice Table 4: 2-Transitive Representations of Almost Simple Groups. We will not need an explicit list of the affine 2-transitive groups. Also, all primitive rank 3 groups are classified (again using the classification of the finite simple groups). They fall into three classes: the almost simple case, the affine case and the grid case. We briefly describe the results in each of the three cases. CLASS I. The Almost Simple Case. First, in order not to mention the same permutation representation twice, we make the four classes of finite simple groups two by two disjoint by deleting those groups in a class that already appeared in a previous class, e.g. we remove L 2 (4) = L 2 (5) from class 2 because this is isomorphic to A 5 in class 1. The classification has been achieved by various people for the respective classes of simple groups: Bannai [3] for the alternating groups (table 5), Kantor & Liebler [59] for the classical Chevalley groups (tables 6 and 7), Liebeck & Saxl [67] for the exceptional Chevalley groups and the sporadic groups (tables 6,8 and 9). So tables 5 to 9 contain all the rank 3 representations of almost simple type. As a general rule, we always list the smallest possible group G; other groups are obtained by adjoining automorphisms 16

17 of G (provided this larger group still acts on Ω). For some classes, we also list the pointstabilizer (denoted by G x ). If not, we list the set Ω by writing a typical element, the full set is obtained by taking the orbit of the typical element under G. G Ω v k;l λ;µ n(n 1) (AL1) A n pair in n-set 2(n 2); (n 2)(n 3) n 2;4 2 2 (AL2) A 8 line in P G(3, 2) 35 16;18 6;8 (AL3) A 9 P G(1, 8) in 9-set ;63 28;24 (AL4) A splitting of 10 points ;100 8;4 Table 5: Rank 3 Representations of Alternating Groups. In order to decide whether an element σ in Aut(G) extends the rank 3 representation, simply let it act on a typical element; if this is possible and the result is inside Ω, then {σ, G} generates a larger rank 3 group on the same set Ω. We illustrate this with an example that we will need anyway later on: consider example (AL3) (of table 5). There are in total 240 P G(1, 8) s in a set of 9 elements and A 9 acts in two orbits twice on 120 of them. So an element of S 9 \ A 9 interchanges these two orbits and hence S 9 does not act as a rank 3 group on 120 points. But in example (AL1), S n does act on the set of duads and hence this gives us a rank 3 representation. In some of the tables, we also list the parameters (v, k, l, λ, µ) of the corresponding strongly regular graphs (see subsection 2.2 for the definitions). These will play a crucial role in our proof. The parameters of the complementary strongly regular graph can be computed easily (see also subsection 2.2) and are not always included in the tables. In the proof of the main result, we will however use these values without further reference. However, the parameters of the strongly regular graphs related to the groups of table 6 can be found in subsection 4.2. CLASS II. The Affine Case. The complete classification of this class is due to Liebeck [65]. He subdivides this class into three subclasses. We give a very brief description in all of these cases. Let us fix our notation: here Ω is an affine space V n (q) of dimension n over GF (q). We denote by G o the stabilizer in G of the zero-vector and we choose n minimal with respect to the property G o ΓL n (q) (as in Liebeck [65]. (A) INFINITE CLASSES. There are 11 infinite classes and we list them in table 10, where we emphasize the geometric properties of G o ; the exact shape of G o (as a group) will be given later if necessary. In most of the cases though, the geometric description suffices (since our proof is using rather geometric arguments). (B) EXTRASPECIAL CLASS. 17

18 G Ω v remark (q n+1 1)(q n 1) (CH1) L n (q) lines in P G(n 1, q) (q+1)(q 1) 2 n 4 q (CH2) S 2n (q) (isotropic) points of P G(2n 1, q) 1 q 1 n 2 q (CH3) O 2n+1 (q) singular points in P G(2n, q) 1 q 1 n 2 (CH4) O 2n(q) + (q singular points in P G(2n 1, q) 1)(q n 1 +1) q 1 n 3 (CH5) O2n(q) (q singular points in P G(2n 1, q) +1)(q n 1 1) q 1 n 3 (CH6) O 10(q) + (q singular 4-spaces in P G(9, q) 1)(q 3 +1) q 1 (CH7) O 2n(2) + non-singular points in P G(2n 1, 2) 2 2n n 1 n 3 (CH8) O2n(2) non-singular points in P G(2n 1, 2) 2 2n 1 2 n 1 n 3 3 (CH9) O 2n+1 (3) points inside a quadric in P G(2n, 3) (3 n +1) n (CH10) O 2n+1 (3) points outside a quadric in P G(2n, 3) (3 n 1) n 2 2 (CH11) O 2n(3) + 3 non-singular points P G(2n 1, 3) (3 n 1) 2 n 3 (CH12) O2n(3) 3 non-singular points P G(2n 1, 3) (3 n +1) n 3 2 (CH13) O 2n+1 (4) non-singular hyperplanes on P G(2n, 4) 2 2n 1 (2 2n 1) n 2 (one orbit) (CH14) O 2n+1 (4) non-singular hyperplanes on P G(2n, 4) 2 2n 1 (2 2n + 1) n 2 (other orbit) (CH15) O 2n+1 (8) : 3 non-singular hyperplanes on P G(2n, 8) 2 3n 1 (2 3n 1) n 2 (one orbit) (CH16) O 2n+1 (8) : 3 non-singular hyperplanes on P G(2n, 8) 2 3n 1 (2 3n + 1) n 2 (other orbit) (CH17) U 2n+1 (q) singular points in P G(2n, q 2 (q ) 1)(q 2n+1 +1) q 2 1 n 1 (CH18) U 2n (q) singular points in P G(2n 1, q 2 (q ) 1)(q 2n 1 +1) n 2 q 2 1 (CH19) U 5 (q) singular lines in P G(4, q 2 ) (q 5 + 1)(q 3 + 1) 2 (CH20) U 2n+1 (2) non-singular points in P G(2n, 4) (2 2n+1 +1) 3 n 2 2 (CH21) U 2n (2) non-singular points in P G(2n 1, 4) (2 2n 1) 3 n 2 (CH22) E 6 (q) points of a building (q 12 1)(q 9 1) q 4 1)(q 1) Table 6: Rank 3 Representations of Chevalley Groups: Infinite Classes. 18

19 G G x v k;l λ;µ λ;µ (CG1) L 2 (8) : 3 7 : ;21 7;4 10;15 (CG2) L 3 (4) A ;45 0;2 36;36 (CG3) S 6 (2) G 2 (2) ;63 28;24 30;36 (CG4) O 7 (3) G 2 (3) ; ; ;504 (CG5) U 3 (3) : 2 L 3 (2) : ;21 4;6 12;12 (CG6) U 3 (5) A ;42 0;1 35;36 (CG7) U 4 (3) L 3 (4) ;105 10;24 72;60 (CG8) U 6 (2) U 4 (3) : ; ; ;520 Table 7: Rank 3 Representations of Classical Groups: Exceptional Classes. G G x v k;l λ;µ λ;µ (EG1) G 2 (3) G 2 (2) ;224 45;45 142;144 (EG2) G 2 (4) J ;315 36;20 234;252 (EG3) G 2 (4) U 3 (4) : ; ; ;528 (EG4) G 2 (8) : 3 ΓU 3 (8) : ; ; ;73920 Table 8: Rank 3 Representations of Exceptional Groups: Exceptional Classes. G G x v k;l λ;µ λ;µ (SP1) M 11 M ;36 9;4 21;28 (SP2) M 12 M ;45 10;4 28;36 (SP3) M A ;60 0;4 47;45 (SP4) M 22 A ;105 18;34 68;54 (SP5) M 23 M ;210 21;4 171;190 (SP6) M A ;140 36;60 87;65 (SP7) M 24 M ;231 22;4 190;210 (SP8) M 24 M ; ; ;504 (SP9) J 2 U 3 (3) ;63 14;12 38;42 (SP10) HS M ;77 0;6 60;56 (SP11) McL U 4 (3) ;162 30;56 105;81 (SP12) Suz G 2 (4) ; ; ;1050 (SP13) Co 2 U 6 (2) ; ; ;896 (SP14) Ru 2 F 4 (2) ; ; ;1280 (SP15) F i 22 2.U 6 (2) ; ; ;2304 (SP16) F i 22 Ω 7 (3) ; ; ;8680 (SP17) F i 23 2.F i ; ; ;25344 (SP18) F i 23 P Ω + 8 (3).S ; ; ;86800 (SP19) F i 24 F i ; ; ; Table 9: Rank 3 Representations of Sporadic Groups. 19

20 n G o (AI1) 1 (AI2) 2m stabilizes direct sum V 2m (q) = V m (q) V m(q). (AI3) 2m stabilizes tensor product V 2m (q) = V m (q) V 2 (q). (AI4) n stabilizes a subspace over GF ( q). (AI5) 2 stabilizes a subspace over GF ( 3 q). (AI6) n stabilizes a non-degenerate Hermitian form in AG(n, q 2 ). (AI7) 2m (AI8) 10 stabilizes a non-degenerate quadratic form of type O2m, ɛ ɛ = +1 or 1. stabilizes a wedge-product 2 (V 5 (q)). (AI9) 8 Ω 7 (q).(2, q 1) G o /Z(G o ) (spin representation). (AI10) 16 P Ω + 10(q) G o /Z(G o ) (spin representation). (AI11) 4 stabilizes the Suzuki-Tits ovoid, q = 2 2h+1. Table 10: Affine Rank 3 Groups: Infinite Classes. n q k;l λ;µ comments (AE1) 2 p p = 7, 13, 17, 19, 23, 29, 31, 47. (AE2) ;36 10;12 G o = (AE3) ;48 13;12 G o = (AE4) ;624 31;12 G o = (AE5) ;64 7;2 G o Sp 4 (3) ; l = (AE6) ;48 13;12 G o Sp 4 (3) ; k = 2 16, l = (AE7) ;384 95;90 k = 15 16, l = (AE8) ; ;90 (AE9) ; ;306 G o = ; k = 45 32, l = G (AE10) ;48 13;12 either o Z(G 27 o).3 2 or G o < and G o GL 2 (3) GL 2 (3). Table 11: Affine Rank 3 Groups: Extraspecial Class. Here, G o is the normalizer of an extraspecial group. We deduce the possibilities in table 11 from Liebeck [65] and Foulser [36]. The comment k = a b means that the suborbit of size k consists of a blocks of imprimitivity of size b. (C) EXCEPTIONAL CASES. Here, the group G o /Z(G o ) is always an almost simple group and again, a list is available. We deduce the information we need from Liebeck [65] and Foulser & Kallaher [37] and list it in table 12. Some parameters are also obtained from Brouwer [9]. CLASS III. The Grid Case. Here, we have a simple group S with S S G S o wr2, where S S o Aut(S) and S o acts 2-transitively on a set of n points. So S o is one of the groups in table 4. Here, Ω = n 2. The notation wr means Wreath product. In our case, H wr 2 is isomorphic 20

21 G o /Z(G o ) n q k;l λ;µ (AF1) A 5 2 q = 31, 41, 71, 79 or 89 (AF2) A ;40 19;20 (AF3) A ; ;380 (AF4) A ;45 2;6 (AF5) S ;480 43;30 (AF6) A ;210 16;6 (AF7) A ; ;220 (AF8) A ;135 56;56 (AF9) A ;210 16;6 (AF10) L 2 (17) ;153 38;42 (AF11) L 3 (4) ;504 61;72 (AF12) U 4 (2) ; ;20 (AF13) G 2 (4) ; ;8010 (AF14) M ;220 1;2 (AF15) M ;132 37;60 (AF16) M ; ;36 (AF17) M ; ;264 (AF18) J ; ;600 (AF19) J ; ;3660 (AF20) Suz ; ;8010 Table 12: The Affine Rank 3 Groups: Exceptional cases. 21

22 to (H H) : 2; if H acts on a set V, then H wr 2 acts on V V as follows: H H in the natural way and the outer 2 by switching the two V s. We refer to such a group G as GRID. 3.2 Distance Transitive Graphs. Let Γ be a graph and suppose Γ is regular as a (g, d p, d l )-gon. Then Γ is called a distance regular graph. If (Γ, G) is moreover a point distance transitive (g, d p, d l )-pair, then (Γ, G), or briefly Γ is called a distance transitive graph. The point graph or collinearity graph of a geometry Γ is the graph obtained from Γ by taking as vertices the points of Γ and as edges the pairs of adjacent points. Similarly for the line graph. There are obvious connections between the point (resp. line) transitivity properties of a geometry and the transitivity properties of its point (resp. line) graph and there are equally obvious connections between the transitivity properties of a geometry and its incidence graph. Now note that the point graph, the line graph and the incidence graph of a regular (g, d p, d l )-gon is a distance regular graph (almost by definition). So, in view of the assumptions on the geometries we consider, distance transitive graphs will play an important role in this paper. Actually, a complete classification (which seems within reach, see Brouwer, Cohen & Neumaier [10]) would make our proof much easier. But the classification is not yet complete and so we must handle some cases by methods depending on the properties of the underlying geometry. But several classes of distance transitive graphs are classified and we will take advantage of such results, except at some places where we can prove a stronger result using geometric properties. We now summarize the results on distance transitive graphs. So let Γ be a distance transitive graph with corresponding group G. Some standard parameters are defined: fix vertices x and y at distance i from each other, then 1. a i = Γ i (x) Γ 1 (y) ; 2. b i = Γ i+1 (x) Γ 1 (y) ; 3. c i = Γ i 1 (x) Γ 1 (y) ; 4. k i = Γ i (x). We have the obvious relations: a i + b i + c i = k 1 ; c i+1 k i+1 = k i b i. If we put d p = d, then an intersection array of Γ is defined as (b 0, b 1,..., b d 1 ; c 1, c 2,..., c d ). 22

23 G d v intersection array (DC1) U 3 (3) 3 63 (6,4,4;1,1,3) (DC2) U 3 (3) : (6,4,4;1,1,3) (DC3) U 3 (5) (12,6,5;1,1,4) (DC4) U 3 (4) : (12,10,5;1,1,8) Table 13: The Admissible Distance Transitive Graphs related to Classical Groups. For reasons that will become clear later, we are not interested in distance transitive graphs of diameter d 3 with c 2 2 or a 1 = 0. Neither are we interested in distance transitive graphs for which the group G acts imprimitively on the set of vertices of Γ. Finally, we assume d 3, since d = 2 corresponds to strongly regular graphs and rank 3 groups. We will call an element of the class of remaining distance transitive graphs an admissible distance transitive graph. Suppose G is a group acting distance transitively and primitively on a distance regular graph Γ. By a theorem of Praeger, Saxl & Yokayama [74] (see also van Bon [92]), there are three possibilities: 1. Γ is a Hamming graph, but then c 2 2, see e.g. Brouwer, Cohen & Neumaier [10]. 2. G is of affine type. Again c G is almost simple. So if we restrict to admissible distance transitive graphs, then the only groups that can occur are almost simple groups. Suppose now G is almost simple and G acts primitively and distance transitively on a distance regular graph Γ. Then the following cases occur: CASE I. The Alternating Case. Here, G is of alternating type. A complete classification has been achieved by Ivanov [52] and Liebeck, Praeger & Saxl [66]. If we restrict to admissible distance transitive graphs, then no examples survive. CASE II. The Classical Case. Here, G is a classical Chevalley group. Again, a complete classification has been achieved recently by Praeger, Saxl & van Bon [75], see also Inglis, Liebeck & Saxl [51]. Their result is basically that there are no surprises compared with [10], considering some additional low-dimensional examples listed in Cohen & van Bon [21]. If we restrict to admissible distance transitive graphs then we find the four examples of table